OpenAI presents GPT-5.2, with improvements in precision and reliability in the professional use of AI

OpenAI has presented its new artificial intelligence (AI) model GPT-5.2the most advanced to date for professional tasks with the Instant, Thinking and Pro versions for ChatGPT, which introduce improvements in reasoning, coding and security, ensuring fewer hallucinations and greater protection for mental health.

The company has more than 800 million weekly users on its ‘chatbot’which represent around 70 percent of assistant usage, as well as more than four million developers building on the OpenAI platform.

In this sense, in order to continue offering a technology that drives everything that professionals, teams and companies want to reliably achieve with AI, OpenAI has presented its new cutting-edge model GPT-5.2, less than a month after its predecessor, GPT-5.1.

The technology company refers to this model as “the best to date“for him everyday professional use, both in the ChatGPT wizard and in the API. Thus, this update of the model includes improvements in multiple aspects, which set “a new standard” in reference evaluations.

INSTANT, THINKING Y PRO

As detailed in a statement, the GPT-5.2 family has been integrated into ChatGPT with the versions GPT-5.2 Instant, GPT-5.2 Thinking y GPT-5.2 Pro, offers significant improvements for work and learning. In the case of version Instantoffers greater speed and best answers to information seeking questionsstep-by-step guides, technical writing and translation.

Likewise, GPT-5.2 Instant is also more effective when used to support study and skill developmentas well as to offer clearer work and professional guidance.

On the other hand, with GPT-5.2 ThinkingOpenAI has highlighted the improvements introduced that make it “the best model to date for professionals”. This is because their ability to solve more complex problems with greater precision.

For example, OpenAI has referenced improvements in the way spreadsheets are organized, as well as advances in creating presentations. All this is supported by its results in relevant tests such as GDPvalthe ‘benchmark’ that evaluates 44 real professional tasksincluding writing documents, in addition to the aforementioned tasks, where it achieves scores of “of expert level“consolidating itself as the first model to achieve this.

Specifically in this test, GPT-5.2 Thinking outperforms or equals the best professionals in the industry for all 44 tasks in a 70.9 percent of caseswhile GPT-5 Thinking only achieved this in 38.8 percent of the cases.

Finally, the version GPT-5.2 Pro comes as “the smarter, more reliable model for difficult questions. This option has a higher response time, however, it offers higher quality results. As a consequence, in preliminary tests, it shows Higher accuracy, fewer major errors, and superior performance in complex domains like programming.

PERFORMANCE AND SECURITY

In addition to all this, the new model also stands out in tasks of codingwhere GPT-5.2 Thinking achieves 55.6 percent in SWE-bench Pro, in tasks of scientific reasoningwhich improves results according to GPQA Diamond, and in tasks of mathematicswhere it scores 40.3 percent on FrontierMath.

Another improvement to take into account is that GPT-5.2 Thinking generates fewer alterations than GPT-5.1 ThinkingSpecifically, OpenAI has highlighted that incorrect answers were 38 percent less frequent. As a result, users will be able to use ChatGPT more reliably for research, writing, analysis or decision making in daily work.

Following this line, the advances in security when using the ‘chatbot’with a special focus on the mental health. This is because GPT-5.2 now more accurately detects a user’s possible signs of discomfort and monitors their responses to offer help. This security improvement has been implemented after rejecting responsibility in the case of the death by suicide of a teenager who used ChatGPT as a confidant.

GPT-5.2 TO BUILD AGENTS

On the other hand, OpenAI has detailed that in the API, the new GPT-5.2 with reasoning is presented as the best option for developers at the time of build agents.

This is because it provides significant improvements in general intelligenceunderstanding long texts, autonomous use of tools by the agent and the vision, since reduces error rates in graph reasoning and can more accurately interpret diagrams or how elements are placed within an image, as the company has specified.

In this way, the technology company led by Sam Altman ensures that GPT-5.2 surpasses any of its previous models in the complete execution of complex real-world tasks.

With all this, OpenAI has indicated that it has started deploying GPT-5.2 (Instant, Thinking and Pro) for users subscribed to the Plus, Pro, Business and Enterprise plans. Likewise, the implementation will be carried out gradually.

OpenAI has presented this new model a week after declaring the company in “code red” under pressure from its competitors, especially Google. At that time, Altman detailed that would restructure its efforts to improve the quality of ChatGPTwith improvements in customization functions for users and more speed, before focusing on other products. Likewise, he announced that he would launch a new reasoning model that would surpass Google’s Gemini 3 model.